Research On Facial Expression Recognition Based On Convolutional Neural Network

There are many ways for humans to express emotions,such as gestures,limbs,expressions,etc.Among them,facial expressions are the most intuitive way to express human emotions in human-to-human communication.With the rapid development of computer vision,facial expression recognition technology,as a bridge of human-computer interaction,has attracted the attention and research of more and more scholars.After years of research by scholars,facial expression recognition technology has been successfully applied in many fields,such as humancomputer interaction in robots,remote online education,and medical monitoring.How to effectively extract facial expression features is the key to facial expression recognition technology.In the past,some methods that relied on humans to extract facial expression features were not only time-consuming and labor-intensive,but also heavily relied on the experience of researchers.With the gradual rise of deep learning,because the convolutional neural network has the advantages of automatically extracting facial expression features,fast calculation,and high recognition rate,it is gradually replacing traditional extraction methods.Although convolutional neural networks can automatically extract facial expression features,how to design convolutional neural networks to extract more comprehensive and deeper facial expression features is still the main direction and hotspot of current facial expression recognition research.In response to this problem,this paper uses the VGG16 network model as the basic network for extracting facial expression features,and researches and improves it,so as to improve the network model's ability to recognize facial expressions.The main research work completed in this paper is as follows:1.According to the development trend and subject requirements of facial expression recognition,I have learned relevant facial expression recognition network models and algorithms,and used the VGG16 network model as the basic network model for facial expression feature extraction to study and improve it.This paper designs an improved scheme of the VGG16 network model,which improves the facial expression recognition ability of the changed network model.2.After analyzing the structural characteristics of the VGG16 network model,first replace the first two fully connected layers of the three fully connected layers of the VGG16 network model with convolutional layers to reduce the parameters of the network model,at the same time,add a context-aware pyramid module before the convolutional layer to expand the receptive field of the network model;Then introduce category attention,pay attention to the feature area that is conducive to facial expression recognition,generate category loss function,and add it to the crossentropy loss as the loss function for network model training.The experimental results show that the recognition rates of the improved network model on the facial expression datasets RAF-DB and FERPlus are 86.70% and 89.63%,respectively,the facial expression recognition rates of the original VGG16 network model in these two datasets are 81.68% and 87.28%,respectively,this result shows that the improved network model has a higher recognition rate of facial expressions than the original VGG16 network model.3.For the VGG16 network model,when extracting facial expression features,it is a single network structure mode,which makes the extracted facial expression features insufficient,leading to the problem that the network model has a low recognition rate of facial expressions.This paper uses an iterative feature fusion method to fuse the extracted features of the VGG16 network model from top to bottom to promote the fusion of the contextual information of facial expression features,making the features extracted by the network model more comprehensive and rich,at the same time,the attention mechanism GCT(Gated Channel Transformation for Visual Recognition)is used on the backbone network of the VGG16 network model to further enhance the network model's ability to extract facial expression features,thereby improving the recognition rate of the network model for facial expressions.The experimental results show that the method designed in this paper has a recognition rate of 87.84% and 56.88% on the facial expression datasets RAF-DB and SFEW,respectively,while the original VGG16 network model has a facial expression recognition rate in these two datasets.They are 81.68% and 50.00% respectively,so the improved network model has a higher improvement than the original VGG16 network model.